In article <6t84i1$2es at pmsnnews.best.ms.philips.com> R. Storm,
rstorm at best.ms.philips.com writes:
>I would like to ask someone to take a look at it (form a biological point
>of view), and tell me what else I need to know, or whats wrong in it.
>
Hi Ray,
One comment I could make about your proposed model concerns the "realism"
of treating a neuron as a summing unit with a threshold. This would
really be a very, very rough approximation. Neuronal membrane responses
are highly nonlinear. Also, because of a property called "inactivation"
that the voltage-gated ion channels have, and because the voltage change
is a function of the total membrane conductance, the firing threshold for
a neuron is not exactly a fixed value. Rather, the threshold can move
around depending on what synaptic conductances (for example) are active,
and also depending on the steepness of the voltage trajectory. These are
somewhat complicated effects to simulate on a large scale, because they
usually require solving a bunch of differential equations for _each_
neuron. But there may be some simplifying assumptions that can be made,
while still preserving the important nonlinearities.
Another, and perhaps more important, comment has to do with modelling
synapses. In general, synaptic strength is also not a fixed quantity.
Aside from really gross long-term changes in synaptic efficacy (i.e.,
Long Term Potentiation and Depression), there are changes that act
constantly and at millisecond time scales. For example, after an initial
stimulus, most central synapses will experience either a facilitation or
a depression that lasts for tens of milliseconds. Thus, _each_ synapse
may adjust its strength somewhat depending on the recent history of
activity at that synapses. For more info on this point, and methods for
modelling it, you might want to look up papers by Tsodyks and Markram.
Cheers,
Matt Jones